But your tests were done as one single process. In prod, there are at least 100 or even 1000 queries running simultaneously. I wonder how fast tokudb will be and how much memory it will need?

Excellent post. Very usefull. I hope to see you again at Conf’10.

Cheers from Brazil,

Alex

show variables like ‘toku%’\G
*************************** 1. row ***************************
Variable_name: tokudb_cache_size
Value: 2072520704
*************************** 2. row ***************************
Variable_name: tokudb_checkpointing_period
Value: 60
*************************** 3. row ***************************
Variable_name: tokudb_commit_sync
Value: ON
*************************** 4. row ***************************
Variable_name: tokudb_data_dir
Value:
*************************** 5. row ***************************
Variable_name: tokudb_debug
Value: 0
*************************** 6. row ***************************
Variable_name: tokudb_init_flags
Value: 11141313
*************************** 7. row ***************************
Variable_name: tokudb_log_dir
Value:
*************************** 8. row ***************************
Variable_name: tokudb_max_lock
Value: 8192
*************************** 9. row ***************************
Variable_name: tokudb_version
Value: mysql-5.1.36-2.1.0-13713-tokudb-2.1.0-13713
9 rows in set (0.00 sec)

tokudb_cache_size
tokudb_commit_sync
tokudb_checkpointing_period

are configurable in my.cnf

To answer your question about memory size:

By default, TokuDB uses about half the memory of the machine as buffer pool (or disk cache). That leaves the other half of memory for the OS to allocate. One important use for the OS is the file system cache. Data is stored uncompressed in the TokuDB cache, but is stored compressed in the file system cache. Thus, we like as much memory as we can get for TokuDB, but we like to leave some for the OS. Half seems like a good balance in most cases.

You can adjust the TokuDB memory usage using the tokudb_cache_size server variable.